2

I just watched a couple of videos on ChatGPT:

https://www.youtube.com/watch?v=Fw5ybNwwSbg https://www.youtube.com/watch?v=UpwsQrc2Jls ... and then this: https://www.youtube.com/watch?v=m_aPB2pPzqE

Beyond the obvious, this technology seems to give an advantage to those who can ask the right questions. Has anyone here used it to write code, or answer a question that stumped you?

6
  • Not an answer (especially not in the long run) but related to those who might not be aware: meta.stackoverflow.com/q/421831/3984613
    – Ghanima Mod
    Feb 21, 2023 at 11:08
  • 1
    @Ghanima: Interesting, but banning a new technology doesn't strike me as the "smart move".
    – Seamus
    Feb 21, 2023 at 20:07
  • 1
    Certainly not. They're probably (re)searching the "smart move" right now.
    – Ghanima Mod
    Feb 22, 2023 at 6:51
  • @Seamus agreed, but that opinion isn't popular on Meta.SE meta.stackexchange.com/a/384405/178179 Feb 28, 2023 at 16:00
  • "this technology seems to give an advantage to those who can ask the right question" -> Pretty sure that the technology of language already does this very substantially, and it is certainly the case on SE since the dawn of time.
    – goldilocks Mod
    Mar 8, 2023 at 12:44
  • ChatGPT has just been superseded by GPT-4, which OpenAI admit can create incorrect answers, so, SE is safe for a while yet.
    – Chenmunka
    Mar 15, 2023 at 12:28

2 Answers 2

1

The issue is that we don't know when ChatGPT is correct. Therefore ChatGPT will not render StackExchange obsolete yet (or anytime soon). Generic question-answering is hard. But question-answering systems may assist human in answering, esp. when they provide attribution, which helps humans factcheck (as Bing is already doing but coarsely).

3
  • I'm stumped as to why you called this an answer, but never mind that for now. What prompted my comment was your statement: "The issue is that we don't know when ChatGPT is correct." I don't see why that's relevant because we also don't know when a real person's answer is correct. But maybe I'm mising yr point?
    – Seamus
    Feb 24, 2023 at 0:24
  • @Seamus SE provide votes, edits, accepted answers, rep, etc. that allows humans to review/assess answers. FYI: Integrate ChatGPT network-wide Feb 24, 2023 at 0:44
  • Right, so if you were asking the question outside of an SE context, you have no checks and balances on chatGPT's "creativity", fact wise.
    – goldilocks Mod
    Mar 8, 2023 at 12:42
0

There's a huge misunderstanding of ChatGPT at the moment. And it's this: ChatGPT is a language model. It's not a reasoning engine or an expert system. It can search content from the internet (copied a year or two back) and it can intelligently collate that information.

That's truly amazing. It really is. It takes our understanding of what a language model can do way further forward than many of us were expecting. But it does not "understand" what it's telling you. It can't reason or diagnose, unless discussion of similar symptoms have already occurred somewhere and been documented in a place it can read.

Trivial examples showing it's lack of cognitive ability include it listing a chronological sequence of events where the dates are all out of order. IE: it can't even count.

In almost all technical forums this same question keeps bouncing about and the simple answer is not yet.


In an AI lecture at university (2006) I was told one of the greatest problems in AI is often getting models to interact with one another. Getting a speech recognition model, to communicate with a natural language model might be okay, but then get that model to interact with a logical reasoning engine and image recognition and you suddenly see how incompatible they are.

Wile this has, no doubt been improved upon a lot since 2006 it remains one of the more challenging issues. Asking ChatGPT to diagnose a problem is not so distant from googling it and reading the articles yourself, perhaps a bit less time consuming.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .